林明志、盧曉珍、楊孟翰 簡廷因、陳怡安、孫安承

1. Distributed Algorithms for Constructing Approximate Minimum Spanning Trees in Wireless Sensor NetworksIEEE Transactions on Parallel and Distributed Systems, vol. 20, no. 1, January 2009 林明志、盧曉珍、楊孟翰 簡廷因、陳怡安、孫安承

2. Agenda • Overview • Algorithm and Analysis • Random-NNT • Coordinate-NNT • UDG-NNT • GHS V.S. NNT • Simulation Results • Dynamic Algorithm for NNT • Concluding Remarks and Future Work

3. Overview D97725004 林明志

4. Motivation • The minimum spanning tree (MST) problem is an important and commonly occurring primitive in the design and operation of data and communication networks • Algorithms require large number of messages and time make them impractical for resource-constrained networks such as wireless sensor networks

5. Introduction • The classical GHS algorithm • Essentially optimal to the message complexity • Θ (nlnn + |E|) messages • Time efficient distributed algorithms • Essentially optimal to the time complexity • O (Diam(G) + n1/2polylog(n)) time • Involve a lot of message transfers

6. Introduction (cont’d) • Wireless network modeled by a unit disk graph (UDG) • Ω (nlnn) messages • Sensor network • Local and distributed • Energy-constrained • Dynamic (reconfiguration)

7. Introduction (cont’d) • Develop simple local distributed algorithms which are energy efficient, and preferably also time efficient, even at the cost (quality) of being suboptimal

8. Introduction (cont’d) • There has been work on local algorithms for construction of low-weight connected subgraphs but little work on localized construction of exact or approximate MSTs • A structure is low-weight if its total edge length is within a small factor of the total edge length of the MST, but the structure may have cycles • Local minimum spanning tree (LMST) • Not a low-weight structure

9. Nearest Neighbor Tree (NNT) • A fundamental step in MST problem is cycle detection • NNT: Each node chooses a unique rank, a node connects to the nearest node of higher rank • The NNT scheme is closely related to the Nearest Neighbor algorithm • NNT with any ranking of the nodes gives an O(lnn)-approximate to MST

10. Definition • Weighted MST problem • Find a tree T spanning N nodes such that is minimized • Radiation energy rα • When α = 1, the traditionalMST problem • α is 2 and can range up to 4 in environments

11. Definition (cont’d) • Cost (quality) • e: an edge of T • Work • r: transmission distance for message i • M: number of messages exchanged by the nodes

12. Network model • Wireless network composed of n nodes distributed uniformly at random in a unit square • Local broadcasting, d(u, v)≦R • Complete graph model • UDG, multihop setting • To have a connected graph, WHP ( ) ‧v R u

13. Contributions and Results • Random-NNT • Ranks are chosen randomly • Coordinate-NNT • Ranks are based on coordinates of the nodes • UDG-NNT • Multihop wireless networks modeled by a UDG

14. Contributions and Results (cont’d) • The tree produced by the NNT has low cost • The NNT can be used to design a simple dynamic algorithm for maintaining a low-cost spanning tree • The time, message, and work complexities of the NNT algorithms are close to the optimal in several settings

15. Performance of NNT and GHS • Quality bounds • Co-NNT is better than Random-NNT • Message, time, and work complexity • NNT have lower message, time, and work complexity compared to the message-optimal GHS

16. Performance of NNT and GHS (cont’d) • Simulation results • Work and messages for NNT are significantly smaller than for GHS • Quality of the NNT trees is very close to MST • Maintaining a low-cost tree dynamically • The expected number of rearrangements (nodes whose outgoing edge must change) is O(lnn)

17. A local distributed algorithm for construction of approximate MST F95942045 盧曉珍

18. The NNT algorithm • IDand Rank simple, local, no cycle, and robust

19. Lemma 1 • NE(v): the size of the neighborhood that node v needs to search to find its connecting edge • proof: v U1 u2 ui

20. The NNT Algorithm: Random-NNT D95922001 楊孟翰

21. The NNT Algorithm • Three types of messages: • Request • Available • Connect • For all nodes, these phases are synchronized by making all nodes wait for Ti, the time required to complete exchanging the message, in phase i

22. Algorithm 2: Distributed NNT algorithm for wireless networks • The algorithm is executed by each node u independently and simultaneously • Message format <msg name, sender, [recipient], [other info]>

23. Analysis of the NNT algorithms • The quality of the tree T produced by NNT • The number of messages M • The work complexity • The time complexity of NNT algorithms

24. Random-NNT • Theorem 2. E[Qα(Random-NNT)] • O(lnn) for α=2 • O(n1-α/2) for α<2 • O(1) for α>2 ri=2*ri-1 Unit square u

25. Proof of Theorem 2 Given one node u, 令符合條件的NN是v, v必定出現在Ri的機率 (也就是要到phase i才能找到符合條件的NN v), 對所有i加總, 再對所有u加總

26. Random-NNT (cont’d) • Theorem 3. The expected work complexity of Random-NNT algorithm E[W] • O(lnn) for α=2 • O(n1- α/2) for α<2 • O(1) for α>2 • A(v, u, i) = the event that v replies to u in phase i • Ti = the event that u needs ith transmission (u進入phase i) v是kth NN

27. Proof of Theorem 3 v在Ri v是kth NN 構成Ci-1 Given one node u, 令傳’available’是v (kth NN of u), v在ith phase傳的機率, 對所有k加總, 考慮radiation energy(ri的α次方) rank(u)最大 構成Ci-1 對所有u加總

28. Random-NNT (cont’d) • Corollary 4. for i ≧2 • The expected number of nodes that needs ith transmission: • The expected number of required transmissions by a node to find a higher ranked node: 對所有nodes加總 對所有phases加總

29. Random-NNT (cont’d) • Theorem 5. The expected message complexity of Random-NNT algorithm is O(n) Theorem 3去掉radiation energy (r的α次方)

30. Random-NNT (cont’d) • Scheduling messages and resolving collisions • Ki= number of messages that potentially collide with u’s messages in phase i; ki ≧2 • Fi= a time frame containing at least ki time slots; each phase i contains O(lnn) such time frames • If a message is in collision, u picks a random slot in the next frame for this message • Probability of a message not colliding = (1-1/ki)ki-1 ≧ 1/e • The time to complete phase i = Ti = O(kilnn) 所有time slots都不會collide

31. Random-NNT (cont’d) • Theorem 6. The time complexity of Random-NNT algorithm is O(ln3n) WHP The total number of messages by nodes within r1 (phase 1)

32. Proof of Theorem 6 • Let |Ci|=y and |Ci-1|=x; z=|Ri|=y-x • Assume y≧60lnn=> E[x]≧y/4≧15lnn

33. Proof of Theorem 6 (cont’d) any node在any phase收到O(lnn)messages的機率的期望值

34. Analysis of Coordinate NNT D97922019 簡廷因

35. Analysis of Coordinate NNT • The expected quality • The expected work complexity • The expected message complexity • The time complexity

36. Proof of Theorem 7 • Theorem 7. The expected quality of Co-NNT for and 2 are and , respectively.

37. Proof of Theorem 7 (cont’d) • We put the cell containing u, then one cell from above u and one cell from below u, by interleaving them. • We put the cells in Column k + 1 in their original order beginning from the bottommost cell to the topmost cells , then the cells in Column k + 2 in the same order, and so on.

38. Proof of Theorem 7 (cont’d) • Node u connects to a node in the ith next cell, if the next cells are empty and there is a node in the ith next cell. • The probability that the next cells are empty is 所有其他n-1個點都在這 ith u … v … i-1 n-(i-1)

39. Proof of Theorem 7 (cont’d) • P’ be the probability that there is a node in the ith cell given that the first cells are empty • Pibe the probability that u connects to a node v in the ith cell

40. Proof of Theorem 7 (cont’d)

41. Proof of Theorem 8 • The expected work complexity of Co-NNT algorithm, for and 2 are and , respectively.

42. Proof of Theorem 8 (cont’d) • We subdivide the area into cells and consider the rearrangement of the cells in a single row as described in the proof of Theorem 7. • The transmission radius for ith phase is • Length of each cell is cells cells cells … … i-1 phase i phase

43. Proof of Theorem 8 (cont’d) • A node u need ith transmission if the next cells are empty. • Let Ti be the event that u needs ith transmission. and for

44. Proof of Theorem 8 (cont’d) • The number of available messages u receives in phase i is the number of nodes in cells that are covered by the transmission i but not by transmission . cells cells cells … … i-1 phase i phase

45. Proof of Theorem 8 (cont’d) • For , the expected number of such nodes in these cells, given that the first cells are empty, is • The expected number of replies in the first phase is

46. Proof of Theorem 8 (cont’d) • In addition, in each phase, there are at most one request message and one connect message by u. Thus, the expected work by u

47. Proof of Theorem 8 (cont’d) • The total work by n nodes • Putting and 2, we have the desired result.

48. Proof of Theorem 9 • Theorem 9. The expected message complexity of Co-NNT algorithm is .

49. Proof of Theorem 9 (cont’d) • When , the total work is equal to the number of messages M. Thus, using , we have .

50. Proof of Theorem 10 • Theorem 10. The time complexity of distributed Co-NNT algorithm is O(ln3n) WHP.